In turbulence observation system, noise signal is random and difficult to identify, which will pollute the real signal and affect the quality of the data. To eliminate the noise signal, the article puts forward a kind of adaptive variable step-size de-noising algorithm. Firstly, raw data is changed into corresponding physical parameters, and spectral analysis is used to analyze the relationship among these parameters, and then, according to the correlation to construct the variable step-size de-noising algorithm, and through error to adjust shape of the step size factor to control the optimal weight coefficient. Finally, simulation and observation data is used to verify the effectiveness of the algorithm, and Goodman's filter algorithm is compared with the algorithm. The results show that the algorithm has higher precision and the noise is effectively reduced.